Monthly Archives: August 2016

Terrorists Vs. Chairs: An Outlier Story

The other day I needed to know how many people were killed by chairs, and while searching I came across the Washington Post’s You’re More Likely To Be Fatally Crushed By Furniture Than Killed By A Terrorist. It argues that worrying about terrorism is irrational, because terrorists kill fewer people each year than falling furniture, and nobody cares about that:

Consider, for instance, that since the attacks of Sept. 11, 2001, Americans have been no more likely to die at the hands of terrorists than being crushed to death by unstable televisions and furniture […] What accounts for the fear that terrorism inspires, considering that its actual risk in the United States and other Western countries is so low? The answer lies in basic human psychology.

I once saw the perfect response to this kind of argument on Twitter, but I forgot to screenshot it, so I’ll have to try to draw it from memory here.

One person posted a graph that looked something like this:

And somebody else edited it to look like this:

And whoa I had never realized before how sketchy it is to start your interval for recording the average number of terrorist attacks the day after the last major terrorist attack.

I mean, I know why people do it. It’s because September 11 was an “outlier”, and outliers should not be counted. Problem is, depending on your distribution, sometimes “outliers” are the only thing that matters.

Let me give an example. Suppose I’m trying to make an argument that earthquakes are totally not a problem for Haiti at all, that there’s no need to invest in earthquake preparedness, and that Haitian people who worry about earthquakes are stupid. I make a graph showing that since January 13, 2010, fewer Haitians have died per year from earthquake-related causes than from crazy furniture-related mishaps. This is totally 100% true. Look at those stupid Haitians, worrying about something that on average never hurts anybody!

(the Haitian earthquake of January 12, 2010 killed about 100,000 people)

I’m sure there are a zillion small Richter 1.0 and Richter 2.0 earthquakes in Haiti all the time. I’m sure our monitoring interval of January 13, 2010 to present picked up lots of these and correctly noted that they don’t kill anybody. The only Haitian earthquakes anyone needs to worry about are the outliers.

If you start your monitoring interval on January 13, earthquakes kill 0 people/year. If you start it on January 11, earthquakes kill 20,000 people a year. Neither of these is entirely fair – one is purpose-designed to maximize casualties, the other to minimize it. I don’t think there’s an obvious fair way to do things – the best solution would be extend the interval back to infinity, but then you get into problems like Haiti having fewer people back in the day, or Haiti not having risen out of the sea yet back in 4,000,000,000 BC. Maybe the best solution is to pick an arbitrary block of time like “the last fifty years”, or to do something very complicated like using the remote historical record to produce earthquake numbers and then combine it with modern populations to produce expected casualties.

The same is true of September 11. Start the interval September 12, and you get 5-10 terrorism deaths/year. Start it September 10, and you get 200. I don’t know when the best time to start it would be. If I had to choose something, I would say maybe 1985, when jihadist terrorism got started after the Soviet invasion of Afghanistan. But someone else could choose 1776, or 2000, based on similarly arbitrary criteria. And it would all be irrelevant – September 11 either made terrorists more ambitious, made security forces more watchful, or both, and so probably changed the calculus for good.

Granted, even when you include September 11, terrorism isn’t the worst thing, and people probably do overestimate it. So forget terrorism. On average, the flu kills something like 20,000 people worldwide each year. That’s a lot, but not apocalyptically much. If you go back year after year, the average stays at something like 20,000/year, right up until you get to 1918, when about 100,000,000 people died. So flu deaths over the last century average about 1 million/year. But three years from now, average flu deaths over the last century will average about 20,000 year. A death rate of only 20,000/year might make our current efforts to contain the flu seem excessive compared to other diseases. But a death rate of 1 million/year makes them look if anything the opposite.

Even worse: did you know that giant asteroids kill about a hundred people per year, on average? This is admittedly an odd definition of “kill” and “average” given that no human being has ever been killed by a giant asteroid. But given that giant asteroids strike Earth about every ten million years, and an asteroid strike today might kill about a billion people, on average giant asteroids kill about a hundred people per year.

Actually, un-forget terrorism. I have a friend who is very in favor of the War On Terror, and he argues that the problem with terrorism isn’t the average suicide bomber who kills three people. It isn’t even the 9-11 hijackers who killed three thousand people. It’s the group that steals a nuke and kills three million people. Just as “on average” a hundred people die each year from giant asteroid strikes, maybe “on average” thirty thousand people die each year from nuclear terrorism. All you’d need for this to be true is one nuclear attack per century. And that’s as bad as an average flu season!

The thing about falling furniture is that there’s probably not going to be a furniturepocalypse where suddenly millions of people all perish at once after being struck by a really really big desk. Furniture is constant. Terrorism isn’t. The whole point of black swans is that we pay too much attention to constant risks and ignore the outliers, especially the outliers which outlie so far that they haven’t happened yet. That’s true whether it’s terrorism, earthquakes, pandemics, or AI.

I worry that someday many years from now, terrorists are going to have some improbable victory which is even more destructive than September 11. I worry that uncounted people are going to die. And I worry that ten years later, someone is going to post on Facebook about how “From the day after ISIS nuked London through today, on average fewer people per year have died of terrorism than from hilarious accidents involving bedside dressers!”